An interesting look - Mike always writes compelling stuff!
A few things that stand out as important outliers to the analysis, IMO:
- The big one: the summer steam sale, which is where back catalog dictates the vast majority of transactions on Steam (and in my experience, for PC games sales in the west full stop), was still finishing up, as it ended July 9. This has a MASSIVE impact on new game releases (which was the dataset reviewed for this analysis), as it has substantial negative effects on 1st party discoverability (aka "the front page of the store"), purchase behaviors (aka "why spend $20 on one new game, when I can spend $20 on four games that are critical darlings?) and the types of games released (aka "a lot of veteran developers and publishers know releasing new games in this window is a little dicey because of the before statements, so they choose not launch"). I don't think this fully discredits the conclusions or the methodology, but it definitely is a massive consideration towards launch & shelf revenue models for a title.
- Because of the above date selected & summer sale implications, it also makes the YoY comparison double dangerous. In 2018, Steam Summer sale started much earlier (June 28th-July 4th), so comparing revenue and sale rate/attach rate has outliers on both sides, with both scenarios having significant depression due to market conditions (again, this is an opinion that would require actual scrutiny!).
- The "algorithm" for calculating year one revenue based on web-published based public data is super duper shaky in my experience. It's good enough for broad strokes (which is why I think this analysis is compelling as is, and not wrong or bad), but there are a litany of reasons why using that to dictate a business plan level conclusion is not ideal. Steam Spy-esque looks are great for high level scans, but it'd be much more ideal to review a Superdata or typical DMP type of conversion schema to get an understanding of actual unit/revenue result. Since most companies don't have that kind of resource, I don't think this is as big a deal as others might.
- July-August is typically a pretty "slow" 30d window for games sales, compared to Fall/Winter, or a 1Y look (which is tricky to do with the available public datasets, granted). Still good enough for a representative sample, but that seems like a massive caveat.
- The price point data slices are a good practical abstract, but due to the 700+ titles excluded from the analysis via the >10 reviews filter, I'd be more pessimistic towards all titles analyzed posted at the $10 price point, as contrary to popular belief many many many of those 700 games aren't just auto-generated shovelware: they are legitimate "games" from actual people. They might not (and 95%, don't) have the production quality of non-shovelware attempts, but a lot of indie developers (think students, daydreamers, game jammers) start here.
- Removing outliers to assess consistency is worthwhile, but I think that's a dangerous boundary to set for dictating health on PC games sales, as most experienced people in games publishing will tell you that games business is by and large hit driven. That said, if I'm a smaller PC developer, I'd be a fucking idiot to plan my runway length on outlier performance, so this is a small quibble - but I would personally strongly recommend re-including those top and bottom 5% revenue generators for all titles above the $30 price point to provide a better understanding of the premise: how well PC games sell on Steam.
- Finally, and I personally haven't ever done big business analysis towards this, but how the ritual of Early Access for game "preview launches" for more established developers has significant ramifications for a 30d game launch window, both in the data ingested in Mike's analysis (I'm not 100% sure, but I believe there's no way currently via Steamworks API or Steam Spy API to slice out Early Access releases/launches from standard atypical releases/launches) and generally selling a game in the year of our lord 2019. Sometimes the most successful (non-outlier) sellers have slow AF EA launches, and build a community that in turn makes a massive "actual launch". A lot of titles Early Access, especially again from vet studios, so I think this is a pretty important caveat to consider if you're trying to assess how you should sell a game on Steam.
Again, I think all that considered, it is an interesting and worthwhile look at how Steam operates, but I wouldn't go too alarm bells on it's conclusion driven. My takeaway from reading it is more questions than answers, but are still "what is the effect of titles that are marketed vs. not for smaller titles/studios and Steam?", and "how much of competitive problems on Steam are due to algorithm discovery vs. exponentially growing competition (historical and current)?".
Oh, and I think worrying about subscription services on games sales is a false flag right now - but might be an actual thing to consider by next year.